[PATCH] mm: percpu: use kmemleak_ignore_phys() instead of kmemleak_free()

[PATCH] mm: percpu: use kmemleak_ignore_phys() instead of kmemleak_free()

[Patrick Wang]

Kmemleak recently added a rbtree to store the objects
allocted with physical address. Those objects can't be
freed with kmemleak_free(). Use kmemleak_ignore_phys()
instead of kmemleak_free() for those objects.

Signed-off-by: Patrick Wang <patrick.wang.shcn@gmail.com>
---
Similar to:
https://lkml.kernel.org/r/20220628113714.7792-2-yee.lee@mediatek.com

 mm/percpu.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/mm/percpu.c b/mm/percpu.c
index 3633eeefaa0d..27697b2429c2 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -3104,7 +3104,7 @@ int __init pcpu_embed_first_chunk(size_t reserved_size, size_t dyn_size,
 			goto out_free_areas;
 		}
 		/* kmemleak tracks the percpu allocations separately */
-		kmemleak_free(ptr);
+		kmemleak_ignore_phys(__pa(ptr));
 		areas[group] = ptr;
 
 		base = min(ptr, base);
@@ -3304,7 +3304,7 @@ int __init pcpu_page_first_chunk(size_t reserved_size, pcpu_fc_cpu_to_node_fn_t
 				goto enomem;
 			}
 			/* kmemleak tracks the percpu allocations separately */
-			kmemleak_free(ptr);
+			kmemleak_ignore_phys(__pa(ptr));
 			pages[j++] = virt_to_page(ptr);
 		}
 	}
@@ -3417,7 +3417,7 @@ void __init setup_per_cpu_areas(void)
 	if (!ai || !fc)
 		panic("Failed to allocate memory for percpu areas.");
 	/* kmemleak tracks the percpu allocations separately */
-	kmemleak_free(fc);
+	kmemleak_ignore_phys(__pa(fc));
 
 	ai->dyn_size = unit_size;
 	ai->unit_size = unit_size;
-- 
2.25.1

Re: [PATCH] mm: percpu: use kmemleak_ignore_phys() instead of kmemleak_free()

[Andrew Morton]

On Tue,  5 Jul 2022 19:31:58 +0800 Patrick Wang <patrick.wang.shcn@gmail.com> wrote:

> Kmemleak recently added a rbtree to store the objects
> allocted with physical address. Those objects can't be
> freed with kmemleak_free(). Use kmemleak_ignore_phys()
> instead of kmemleak_free() for those objects.

Thanks.  What are the user-visible runtime effects of this?

And are we able to identify a commit for the Fixes: line?

Re: [PATCH] mm: percpu: use kmemleak_ignore_phys() instead of kmemleak_free()

[patrick wang]

On Wed, Jul 6, 2022 at 5:20 AM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Tue,  5 Jul 2022 19:31:58 +0800 Patrick Wang <patrick.wang.shcn@gmail.com> wrote:
>
> > Kmemleak recently added a rbtree to store the objects
> > allocted with physical address. Those objects can't be
> > freed with kmemleak_free(). Use kmemleak_ignore_phys()
> > instead of kmemleak_free() for those objects.
>
> Thanks.  What are the user-visible runtime effects of this?

According to the comments, percpu allocations are tracked
by kmemleak separately. Kmemleak_free() was used to avoid
the unnecessary tracking. If kmemleak_free() fails, those
objects would be scanned by kmemleak, which is unnecessary
but shouldn't lead to other effects.

I didn't observe any anomaly without this commit on riscv
and arm64.

>
> And are we able to identify a commit for the Fixes: line?

0c24e061196c (mm: kmemleak: add rbtree and store physical
address for objects allocated with PA)
Current in mm-stable.

Thanks,
Patrick

Re: [PATCH] mm: percpu: use kmemleak_ignore_phys() instead of kmemleak_free()

[Catalin Marinas]

On Wed, Jul 06, 2022 at 10:44:11PM +0800, patrick wang wrote:
> On Wed, Jul 6, 2022 at 5:20 AM Andrew Morton <akpm@linux-foundation.org> wrote:
> >
> > On Tue,  5 Jul 2022 19:31:58 +0800 Patrick Wang <patrick.wang.shcn@gmail.com> wrote:
> >
> > > Kmemleak recently added a rbtree to store the objects
> > > allocted with physical address. Those objects can't be
> > > freed with kmemleak_free(). Use kmemleak_ignore_phys()
> > > instead of kmemleak_free() for those objects.
> >
> > Thanks.  What are the user-visible runtime effects of this?
> 
> According to the comments, percpu allocations are tracked
> by kmemleak separately. Kmemleak_free() was used to avoid
> the unnecessary tracking. If kmemleak_free() fails, those
> objects would be scanned by kmemleak, which is unnecessary
> but shouldn't lead to other effects.
> 
> I didn't observe any anomaly without this commit on riscv
> and arm64.

What could happen is an increased rate of false negatives as it scans
more than necessary.

> > And are we able to identify a commit for the Fixes: line?
> 
> 0c24e061196c (mm: kmemleak: add rbtree and store physical
> address for objects allocated with PA)
> Current in mm-stable.

I think we could add a Fixes line for the above. For the patch:

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>